I am interested in using Zed X (or the 2i), at least 4, in a ceiling mounted fashion, located in corners of a 7 x 7 x 5 meter capture volume i.e. robot playground. The tentative plan is to use a 4 camera setup connected to an edge PC running ZED SDK-based application for either spatial mapping, object detection, volume estimation, etc and pull lightweight processed data streams into deployment computers orchestrating robot systems.
I have questions about the choice of camera and edge connectivity:
For this application is there a big difference in capabilities of ZED X and ZED 2i other than their connection interface? Is one preferred over the other for indoor/warehouse kind of setting?
Regarding the hardware and connectivity:
2.a. From computational/communication bottleneck perspective, how do I choose the capture card/ZED box quantity. For instance, since each capture card port can plug to a 1→4 GSML2 connector, how many cameras can one ZED box and a quad capture card run?
2.b. For a combination of 4x ZED X and 2x ZED One (for 2D streams), is it recommended to separate ZED box or one box and a quad capture card suffices?
Thank you for the detailed description of your setup; a fixed, ceiling mounted multi camera volume is a use case we support well, so let me answer point by point.
The interface is the main difference, but for your installation it is exactly the difference that matters most. GMSL2 carries video and power on a single coaxial cable (PoC), and it is qualified up to 15 m per camera, so four corner mounted cameras at 5 m height are reachable with a single cable run each. With USB3 you are practically limited to the cable supplied with the camera and you would need active extenders, which are a common source of instability in permanent installations.
Beyond the interface, the relevant differences for you are:
Shutter: the ZED X uses a global shutter, the ZED 2i uses a rolling shutter. If the robots in the playground move at non trivial speed, the global shutter avoids geometric distortion in the depth maps.
Hardware synchronization: GMSL2 cameras can be frame synchronized on the capture card, which is important when you fuse four viewpoints of the same moving object. USB3 cameras cannot.
Enclosure: ZED X is IP67, ZED 2i is IP66.
Sensors: ZED 2i has a full 9 DoF stack (IMU, magnetometer, barometer); ZED X has an IMU. For statically mounted cameras this is not a discriminant.
About optics, from a corner at 5 m the far corner of a 7 x 7 m area is roughly 11 m away, so the lens choice is a trade off between coverage and range. The 2.2 mm version gives you the wide field of view needed to see the whole volume from one corner; the 4 mm version gives better depth quality at distance but narrower coverage. The figures per model and lens are in the datasheets, and the depth range parameters are documented in Depth Settings and Depth Modes.
Small clarification first, because it changes the arithmetic: the 1 to 4 Fakra cable is a fan out of one physical connector into 4 GMSL2 wires, and each wire hosts one camera. It is not a 1 to 4 multiplier per port.
The ZED Link Quad exposes 8 GMSL2 ports through 2 Fakra 4x1 connectors, organized in 6 groups. The number of cameras and the resolution/framerate you obtain depend on the camera type and the group they are connected to; the authoritative table is in Multi-camera configurations. Two points from that table are relevant to you: a stereo camera alone in a group runs at HD1200@60fps, two stereo cameras in the same group drop to HD1200@30fps, and maximum resolution/framerate requires no more than 2 cameras per 4 wire Fakra connector.
The Quad is supported on AGX Orin and Xavier AGX Developer Kits only; see the compatibility tables in ZED Link capture cards.
The ZED Box Orin with 4 wire GMSL2 connectivity provides 4 GMSL2 ports in 2 groups, so it hosts up to 4 ZED X at HD1200@30fps, or 2 ZED X at HD1200@60fps. Details here: Using GMSL with ZED Box Orin. Please note that a ZED Box Orin cannot be opened to install a different capture card afterwards, the capture card must be chosen at purchase time.
Two more things worth checking while sizing:
Frame synchronization: on the Quad the two deserializers are not synchronized by default. Follow Enable full onboard synchronization to get 15 µs frame accuracy across all four cameras.
In practice the GMSL2 link is rarely the limiting factor; the GPU is. Four simultaneous NEURAL depth instances plus object detection is a heavy load on a single Orin, so plan the depth mode and resolution per camera accordingly, and check the requirements in ZED SDK Specifications.
I recommend splitting it across two hosts. Four ZED X plus two ZED X One means 10 image streams and 6 devices distributed over only 2 Fakra connectors, which breaks the “no more than 2 cameras per 4 wire connector” condition for maximum resolution and framerate, and it concentrates all the depth and AI processing on a single module.
A configuration that keeps every camera at full rate is:
1x AGX Orin Developer Kit with a ZED Link Quad for the 4x ZED X, two cameras per Fakra connector;
Note also that the ZED X One 4K is limited to 1 camera per group and only on the even ports, while the ZED X One GS/S allows up to 2 per group; the ZED X One documentation and the tables linked above cover this.
For the application layer, the pieces you will want are:
Fusion API: handles time synchronization and geometric calibration across cameras, and supports both a local workflow and a network workflow with one publisher per host, which is exactly the two box topology above.
ZED360: computes the extrinsic calibration of the fixed multi camera rig, which is the prerequisite for any digital twin in a common world frame.